
import os
os.environ["OMP_NUM_THREADS"] = '5' # set to avoid memory leak warning when using k-means classifier model

import pandas as pd
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report, ConfusionMatrixDisplay
from sklearn.preprocessing import MinMaxScaler
from sklearn.feature_selection import RFE
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.cluster import KMeans

# Visualisation libraries
import matplotlib.pyplot as plt
import plotly.express as px


df = pd.read_csv("./gym_churn_us.csv")


df.sample(5)


df.info() # no null values

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4000 entries, 0 to 3999
Data columns (total 14 columns):
 #   Column                             Non-Null Count  Dtype  
---  ------                             --------------  -----  
 0   gender                             4000 non-null   int64  
 1   Near_Location                      4000 non-null   int64  
 2   Partner                            4000 non-null   int64  
 3   Promo_friends                      4000 non-null   int64  
 4   Phone                              4000 non-null   int64  
 5   Contract_period                    4000 non-null   int64  
 6   Group_visits                       4000 non-null   int64  
 7   Age                                4000 non-null   int64  
 8   Avg_additional_charges_total       4000 non-null   float64
 9   Month_to_end_contract              4000 non-null   float64
 10  Lifetime                           4000 non-null   int64  
 11  Avg_class_frequency_total          4000 non-null   float64
 12  Avg_class_frequency_current_month  4000 non-null   float64
 13  Churn                              4000 non-null   int64  
dtypes: float64(4), int64(10)
memory usage: 437.6 KB


# Check if there are any rows where months to end contract > contract period
df[df['Month_to_end_contract'] > df['Contract_period']]


# Rename some columns
df = df.rename(columns={
    'Month_to_end_contract': 'Num_months_to_contract_end', 
    'Avg_class_frequency_total': 'Avg_weekly_visits_total', # average frequency of visits per week over the customer's lifetime
    'Avg_class_frequency_current_month': 'Avg_weekly_visits_preceding_month'# average frequency of visits per week over the preceding month
})


df.describe(include='all')


df['Contract_period'].value_counts()

1     2207
12     960
6      833
Name: Contract_period, dtype: int64


X = df.drop('Churn', axis=1)
y = df['Churn']


def preprocess_data(X):
    # for numerical columns
    # ---------------------
    # we normalize using MinMaxScaler to constrain values between 0 and 1

    scaler = MinMaxScaler(feature_range = (0,1))
    df_numeric = X.select_dtypes(include=[np.number])
    numeric_cols = df_numeric.columns.values

    for col in numeric_cols:
        # fit_transform() of scaler can be applied to each column individually
        X[col] = scaler.fit_transform(X[[col]])
    
    return X


X = preprocess_data(X)
X.describe()


# Split feature and label sets to train and data sets - 80-20
# random_state is desirable for reproducibility
# stratify - to keep the same proportion of (yes and no) as input dataset

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 10, stratify = y)
print(X_train.shape, X_test.shape, y_train.shape, y_test.shape)

(3200, 13) (800, 13) (3200,) (800,)


# create a logistic regression classifier
# ---------------------------------------
logreg = LogisticRegression(solver='liblinear', random_state=7)
logreg.fit(X_train, y_train)

# Apply the model on the test data to get a prediction
# ----------------------------------------------------
y_pred = logreg.predict(X_test)
y_pred.shape

(800,)


n = 10
print(f"Probability of first {n} users not churning: \n {logreg.predict_proba(X_test)[:,0][:n]}", end='\n\n') # probability of not churning
print(f"Probability of first {n} users churning: \n {logreg.predict_proba(X_test)[:,1][:n]}", end='\n\n') # probability of churning
print(f"Whether first {n} users churned", y_pred[:n]) # using default threshold of 0.5 to decide if user churn (probability > 0.5) or not (probability < 0.5)

Probability of first 10 users not churning: 
 [0.80407839 0.89458534 0.3542922  0.97464291 0.09258459 0.85284717
 0.73054158 0.82518119 0.0756298  0.01948577]

Probability of first 10 users churning: 
 [0.19592161 0.10541466 0.6457078  0.02535709 0.90741541 0.14715283
 0.26945842 0.17481881 0.9243702  0.98051423]

Whether first 10 users churned [0 0 1 0 1 0 0 0 1 1]


# Get accuracy of the model
# -------------------------
print('Testing accuracy %s' % accuracy_score(y_test, y_pred))

# compute confusion matrix
# ------------------------
cm = confusion_matrix(y_test, y_pred)

disp = ConfusionMatrixDisplay(confusion_matrix=cm)
disp.plot()

print(classification_report(y_test, y_pred))

Testing accuracy 0.9175
              precision    recall  f1-score   support

           0       0.92      0.97      0.95       588
           1       0.90      0.77      0.83       212

    accuracy                           0.92       800
   macro avg       0.91      0.87      0.89       800
weighted avg       0.92      0.92      0.92       800


tn, fp, fn, tp = cm[0][0], cm[0][1], cm[1][0], cm[1][1]
# Precision: Given model predict positive/negative, how good ('precise') is the model in predicting positive/negative
precision_no_churn = tn / (tn + fn)
print(f'precision_no_churn is {precision_no_churn:.3f}')

precision_churn = tp / (tp + fp)
print(f'precision_churn is {precision_churn:.3f}')

#Recall: Given actual is positive/negative, how good is the model in predicting ('recalling') positive/negative
recall_no_churn = tn / (tn + fp)
print(f'recall_no_churn is {recall_no_churn:.3f}')

recall_churn = tp / (tp + fn)
print(f'recall_churn is {recall_churn:.3f}')

precision_no_churn is 0.922
precision_churn is 0.901
recall_no_churn is 0.969
recall_churn is 0.774


### Possible evaluation with another classificaion model e.g. decision tree if using model to predict if new customers will churn or not


n = 6
rfe = RFE(logreg, n_features_to_select=n)
fit = rfe.fit(X, y)

print(f'Number features: {fit.n_features_}')
print("Selected Features: %s" % fit.support_)
print("Feature Ranking: %s" % fit.ranking_)

Number features: 6
Selected Features: [False False False False False  True False  True  True False  True  True
  True]
Feature Ranking: [7 6 5 4 8 1 3 1 1 2 1 1 1]


selected = fit.support_
feature_names = X.columns[selected]
feature_names

Index(['Contract_period', 'Age', 'Avg_additional_charges_total', 'Lifetime',
       'Avg_weekly_visits_total', 'Avg_weekly_visits_preceding_month'],
      dtype='object')


for i in feature_names:
    print(i)

Contract_period
Age
Avg_additional_charges_total
Lifetime
Avg_weekly_visits_total
Avg_weekly_visits_preceding_month


churn_df = df[df['Churn'] == 1].copy()
# churn_df.describe()
churn_features_df = churn_df[feature_names].copy()
churn_features_df = preprocess_data(churn_features_df)
churn_features_df.describe()


SSE = []
for i in range(1,6):
    KMeans_model = KMeans(n_clusters=i, random_state=42, n_init=10, max_iter=300)
    KMeans_model.fit(churn_features_df)
    SSE.append(KMeans_model.inertia_)


# Plot the SSE to decide the n_clusters to use based on elbow plot
labels = {'x': 'No. of Clusters', 'y': 'SSE'}
title = 'Sum of Squared Errors (SSE) to Centroid Based on No. of Clusters'
fig = px.line(x=range(1,6), y=SSE, labels=labels, title=title)
fig.show()


KMeans_model = KMeans(n_clusters=3, n_init=10, max_iter=300, random_state=42)
KMeans_model.fit(churn_features_df)

KMeans(n_clusters=3, n_init=10, random_state=42)

KMeans(n_clusters=3, n_init=10, random_state=42)


user_profile_df = pd.DataFrame(data=KMeans_model.cluster_centers_, columns=churn_features_df.columns)
user_profile_df


labels={'x': 'User Profile'}
heatmap = px.imshow(user_profile_df.T, x=user_profile_df.index, y=user_profile_df.columns,
                    labels=labels, color_continuous_scale='Viridis', title='Heatmap of Churn User Profiles')
heatmap.show()


cluster_labels = KMeans_model.labels_ # cluster label for each user that churn
cluster_labels

array([2, 1, 2, ..., 2, 2, 2])


churn_df['Cluster'] = cluster_labels
churn_df


cluster_churn_count = churn_df['Cluster'].value_counts()
cluster_churn_count

2    512
1    430
0    119
Name: Cluster, dtype: int64


title="% of customers churned in each cluster"
px.pie(values=cluster_churn_count, names=cluster_churn_count.index, title=title)


churn_df['Contract_utilisation_pct'] = (churn_df['Num_months_to_contract_end'] / churn_df['Contract_period']) * 100
churn_df[['Num_months_to_contract_end', 'Contract_period', 'Contract_utilisation_pct']].head()


fig = px.histogram(churn_df, x="Contract_utilisation_pct",
             color='Cluster', barmode='group',
             height=600, title="Contract Utilisation % for Churn users in each Cluster", )

fig.update_xaxes(title_text='Contract Utilisation %')

fig.show()

	gender	Near_Location	Partner	Promo_friends	Phone	Contract_period	Group_visits	Age	Avg_additional_charges_total	Month_to_end_contract	Lifetime	Avg_class_frequency_total	Avg_class_frequency_current_month
2334	1	1	1	1	1	1	1	33	369.234321	1.0	3	1.674872	1.593047
2806	1	0	1	0	1	1	1	25	227.239408	1.0	3	3.060350	3.118923
875	1	1	0	0	1	1	0	29	49.099452	1.0	6	2.631843	2.656269
3698	0	1	0	0	1	1	0	33	60.202737	1.0	1	1.559161	1.520917
927	1	1	1	1	1	1	0	35	114.347092	1.0	7	3.568346	3.548644

	gender	Near_Location	Partner	Promo_friends	Phone	Contract_period	Group_visits	Age	Avg_additional_charges_total	Num_months_to_contract_end	Lifetime	Avg_weekly_visits_total	Avg_weekly_visits_preceding_month	Churn
count	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000
mean	0.510250	0.845250	0.486750	0.308500	0.903500	4.681250	0.412250	29.184250	146.943728	4.322750	3.724750	1.879020	1.767052	0.265250
std	0.499957	0.361711	0.499887	0.461932	0.295313	4.549706	0.492301	3.258367	96.355602	4.191297	3.749267	0.972245	1.052906	0.441521
min	0.000000	0.000000	0.000000	0.000000	0.000000	1.000000	0.000000	18.000000	0.148205	1.000000	0.000000	0.000000	0.000000	0.000000
25%	0.000000	1.000000	0.000000	0.000000	1.000000	1.000000	0.000000	27.000000	68.868830	1.000000	1.000000	1.180875	0.963003	0.000000
50%	1.000000	1.000000	0.000000	0.000000	1.000000	1.000000	0.000000	29.000000	136.220159	1.000000	3.000000	1.832768	1.719574	0.000000
75%	1.000000	1.000000	1.000000	1.000000	1.000000	6.000000	1.000000	31.000000	210.949625	6.000000	5.000000	2.536078	2.510336	1.000000
max	1.000000	1.000000	1.000000	1.000000	1.000000	12.000000	1.000000	41.000000	552.590740	12.000000	31.000000	6.023668	6.146783	1.000000

	gender	Near_Location	Partner	Promo_friends	Phone	Contract_period	Group_visits	Age	Avg_additional_charges_total	Num_months_to_contract_end	Lifetime	Avg_weekly_visits_total	Avg_weekly_visits_preceding_month
count	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000	4000.000000
mean	0.510250	0.845250	0.486750	0.308500	0.903500	0.334659	0.412250	0.486272	0.265721	0.302068	0.120153	0.311939	0.287476
std	0.499957	0.361711	0.499887	0.461932	0.295313	0.413610	0.492301	0.141668	0.174417	0.381027	0.120944	0.161404	0.171294
min	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
25%	0.000000	1.000000	0.000000	0.000000	1.000000	0.000000	0.000000	0.391304	0.124394	0.000000	0.032258	0.196039	0.156668
50%	1.000000	1.000000	0.000000	0.000000	1.000000	0.000000	0.000000	0.478261	0.246310	0.000000	0.096774	0.304261	0.279752
75%	1.000000	1.000000	1.000000	1.000000	1.000000	0.454545	1.000000	0.565217	0.381581	0.454545	0.161290	0.421019	0.408398
max	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000

	Contract_period	Age	Avg_additional_charges_total	Lifetime	Avg_weekly_visits_total	Avg_weekly_visits_preceding_month
count	1061.000000	1061.000000	1061.000000	1061.000000	1061.000000	1061.000000
mean	0.066233	0.449482	0.270189	0.110064	0.424014	0.295047
std	0.193852	0.144758	0.182649	0.123422	0.199706	0.217564
min	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
25%	0.000000	0.350000	0.118671	0.000000	0.290565	0.119013
50%	0.000000	0.450000	0.243699	0.111111	0.428669	0.276658
75%	0.000000	0.550000	0.388984	0.111111	0.562414	0.448716
max	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000

	Contract_period	Age	Avg_additional_charges_total	Lifetime	Avg_weekly_visits_total	Avg_weekly_visits_preceding_month
0	0.559969	0.433193	0.291236	0.120448	0.467211	0.320948
1	0.001055	0.454408	0.271714	0.100541	0.585992	0.480380
2	0.006227	0.449119	0.264000	0.115677	0.277335	0.132697

Installations¶

Dataset description¶

Broad strategy to analyse customer churn data¶

Data Cleaning¶

1. Exploratory Data Analysis (EDA)¶

EDA Findings¶

Most contract period length is 1 month¶

Separate X and y to train logistic regression model for identifying features contributing to churn¶

2. Pre-process data¶

3. Train logistic regression model¶

4. Evaluate the logistic regression model¶

Evaluation¶

1. Select top `n` features that contribute most to churn.¶

2. Group churn users into user profiles¶

No. of clusters (user profiles) to create for those who churn¶

Perform k means clustering using 3 clusters to create 3 user profiles¶

Look at the cluster centroids (centroid values for the churn features) and categorise each cluster into user profiles¶

Interpretation of clusters¶

How much customers are lost in each cluster? - Contributes to deciding which cluster to focus on retaining given a budget constraint¶

Finding potential reasons why customers in these clusters churn¶

Conclusion¶

	gender	Near_Location	Partner	Promo_friends	Phone	Contract_period	Group_visits	Age	Avg_additional_charges_total	Num_months_to_contract_end	Lifetime	Avg_weekly_visits_total	Avg_weekly_visits_preceding_month	Churn	Cluster
7	0	1	0	0	1	1	0	30	217.786641	1.0	0	1.213502	1.049387	1	2
11	1	1	1	0	1	1	0	27	266.783824	1.0	2	2.348535	1.584316	1	1
13	1	1	0	0	1	1	1	27	152.748668	1.0	1	0.000000	0.000000	1	2
21	0	0	0	0	1	1	0	30	153.198594	1.0	3	0.636382	0.000000	1	2
23	1	1	0	0	1	1	1	32	276.079882	1.0	1	1.420637	1.043101	1	2
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3982	1	1	0	0	1	1	0	24	135.395712	1.0	0	1.390251	0.789321	1	2
3984	1	1	0	0	1	6	1	22	101.107544	6.0	2	1.377860	0.343915	1	0
3986	0	1	1	1	1	1	1	34	226.804845	1.0	0	1.399529	0.304130	1	2
3994	1	1	0	0	1	1	0	26	31.066721	1.0	1	1.157451	0.269597	1	2
3996	0	1	0	0	1	1	1	29	68.883764	1.0	1	1.277168	0.292859	1	2

Installations¶

Dataset description¶

Broad strategy to analyse customer churn data¶

Data Cleaning¶

1. Exploratory Data Analysis (EDA)¶

EDA Findings¶

Most contract period length is 1 month¶

Separate X and y to train logistic regression model for identifying features contributing to churn¶

2. Pre-process data¶

3. Train logistic regression model¶

4. Evaluate the logistic regression model¶

Evaluation¶

1. Select top n features that contribute most to churn.¶

2. Group churn users into user profiles¶

No. of clusters (user profiles) to create for those who churn¶

Perform k means clustering using 3 clusters to create 3 user profiles¶

Look at the cluster centroids (centroid values for the churn features) and categorise each cluster into user profiles¶

Interpretation of clusters¶

How much customers are lost in each cluster? - Contributes to deciding which cluster to focus on retaining given a budget constraint¶

Finding potential reasons why customers in these clusters churn¶

Conclusion¶

1. Select top `n` features that contribute most to churn.¶